Dataset · nvidia
Math corpus from Common Crawl
nvidia/Nemotron-CC-Math-v1
133 billion math and code tokens from Common Crawl for LLM pre-training.
What's inside
Nemotron-CC-Math is a large, high-quality math corpus extracted from Common Crawl. According to the card, its volume is 133 billion tokens; the data was used in Nemotron pre-training and preserves valuable math and code information. Metadata lists the size as 100M-1B, while the paper description states 133 billion tokens. The modality is text.
What you can do with it
- Pre-training math LLMs
- Fine-tuning models to solve math and code
- Evaluating the quality of math pre-training data
- Tasks
- Text generation
- Size
- 100M–1B rows
- Downloads this month
- 13.8K
- Likes
- 105
- Access
- on request on Hugging Face
Models trained on it
- nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16Text generation 1.1M
- nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4Text generation 641.7K
- nvidia/NVIDIA-Nemotron-Nano-9B-v2Text generation 204.4K
- nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8Text generation 189.4K
- nvidia/NVIDIA-Nemotron-Nano-9B-v2-FP8Text generation 149.7K
- nvidia/NVIDIA-Nemotron-Nano-9B-v2-NVFP4Text generation 49.8K
- unsloth/Nemotron-3-Nano-30B-A3B-GGUFText generation 31.7K
- lmstudio-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-4bitText generation 30.5K
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/Nemotron-CC-Math-v1")